Machine learning based model for the prediction of dimensionless temperature in multimode heat transfer

Performance prediction is key to engineering analysis of thermal systems. In general, a mathematical model of performance parameter in terms of relevant independent parameters is used for the performance prediction. When the functional relation among the parameters involved is not known a priori , a machine learning based method is a meritorious approach for the evaluation of performance parameter(s). In this study, the dimensionless temperature of a heat generating model with different surface morphologies under varying heat input is the performance parameter of interest, while dimensionless volumetric heat generation, surface roughness and emissivity are the independent parameters. Heat transfer experiments were conducted to generate data and a supervisory machine learning algorithm, namely, Artificial Neural Network (ANN) was used to predict the temperature. A comparison of the temperatures predicted by the ANN model and the empirical correlation developed using least square regression indicated that ANN model fits the dataset significantly better than the regression model. The machine learning model is further analyzed using the Shapley Additive Explanations (SHAP) method from Explainable AI (XAI) to quantify the contribution of each variable to the model prediction. • Efficacy of ANN model to predict the temperature of a heat transfer model is reported. • ANN model fits the dataset significantly better than the regression model. • ANN modelling is effective when function form among the parameters is not known a priori .